See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/238524044

Modeling SMT ferrite beads for SPICE simulation

Article · August 2011

DOI: 10.1109/ISEMC.2011.6038369

CITATIONS READS
8 2,757

3 authors:

C. Rostamzadeh F. Grassi
Bosch Politecnico di Milano
31 PUBLICATIONS   60 CITATIONS    121 PUBLICATIONS   751 CITATIONS   

SEE PROFILE SEE PROFILE

F. Kashefi
University of Texas at Arlington
6 PUBLICATIONS   19 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Radiated susceptibility of wiring harnesses View project

ETOPIA Project - Marie Sklodowska Curie Actions View project

All content following this page was uploaded by F. Kashefi on 20 March 2015.

The user has requested enhancement of the downloaded file.

 Modeling SMT Ferrite Beads for SPICE Simulation
C. Rostamzadeh#, F. Grassi*, F. Kashefi#
#
Robert Bosch LLC - USA, Plymouth, United States
Cyrous.Rostamzadeh@us.bosch.com
*
Politecnico di Milano, Milan, Italy
flavia.grassi@polimi.it
#
Khavaran Institute of Science and Technology, Mashhad, Iran
fred.kashefi@ieee.org

Abstract—A practical approach for lumped-parameter circuit measured and how they can be used for circuit modeling and
modeling of SMT ferrite beads is introduced. Vector Network simulation is often unclear.
Analyzer (VNA) measurements are utilized to provide the To remove the above limitations, the parameters of the
frequency-dependent characteristics for SPICE analysis. The ferrite-bead model need to be considered as frequency-
measured data is imported into the simulation environment via
Analog Behavioral Models (ABM) accounting for the frequency-
dependent quantities [4], to be derived from measurement data
dependent behavior of the ferrite sheets. It demonstrates a [5]. To this end, an ad hoc test setup and a rigorous
significant departure from a simple R-L-C circuit network. methodology based on Vector Network Analyzer (VNA)
Model accuracy is validated by realization of an ad hoc test- measurements is adopted in this paper to extract the SMT
board and by experimental characterization of the component in ferrite bead model-parameters. Model implementation in
terms of Insertion Loss. SPICE is achieved by resorting to Analog Behavioral
Keywords—Surface Mount Technology (SMT), Ferrites, Modeling (ABM) parts, allowing for the inclusion of the
Complex Permeability, PSPICE Analog Behavioral Modeling frequency-dependent characteristics of the ferrite bead into the
(ABM), Insertion Loss (IL). simulation environment [6, 7].
Effectiveness of the proposed modeling approach is
I. INTRODUCTION validated by design and implementation of a PCB test board
Surface-Mount Technology (SMT) ferrite beads as suitable for experimental characterization of the SMT
frequency-selective devices have the potential to solve a large component in terms of Insertion Loss (IL).
spectrum of Electromagnetic Compatibility (EMC) problems.
These components are extensively used in the printed circuit II. SMT FERRITE BEADS: STRUCTURE AND CHARACTERISTICS
board (PCB) applications to mitigate electromagnetic (EM) A. Geometrical layout
noise. In general, EM properties of SMT ferrite beads result in
a high-impedance (resistance) for high-frequency (HF) signals, The geometrical layout of a multi-layer SMT ferrite bead is
thus attenuating HF EM disturbances. The absorbed energy is illustrated in Fig. 1 [1]. The conductive traces (electrodes) are
converted into heat and dissipated by the ferrite. printed on successive ferrite sheets which are stacked up in
Recent innovations of the SMT ferrite bead technology successive layers to achieve high inductance. The electrode
offer an extensive library of components, which enables the pattern on each sheet is connected through a via-hole to
engineers to develop electronic products with most stringent successive layers. The pattern is carefully structured to
EMC requirements. The availability and the economic merits minimize parasitic capacitances between layers (inter-winding
of the PCB SMT ferrite beads offer tremendous benefits for capacitances). The parasitic capacitance is the dominant
the electronic design community. However, in general, EMC element at high frequencies, impacting the behavior of the
engineers are limited with the availability of a reasonably ferrite-bead impedance.
simple and accurate electrical model that could be used for
B. Frequency characteristics and modeling issues
circuit simulation.
Most of the application notes published by SMT ferrite In principle, accurate modeling of the bead-structure in Fig.
suppliers suggest simple series R-L or R-L-C circuit models 1 should require a distributed-parameter approach, where an
[1, 2]. However, this approach results in inaccurate simulation inductance should be associated with each layer and an inter-
of the ferrite’s complex frequency-dependent behavior, as it winding capacitance between adjacent layers [8]. However, it
pertains to value(s) of R and L at a ‘fixed-frequency’, is common practice to resort to the lumped parameter circuit
traditionally at 100 MHz. In a similar fashion, a lumped- network in Fig. 2 to provide an equivalent representation of
element model with constant-valued parameters is proposed in the component at the outer terminals [3, 4, 8].
[3], and it is used to include ferrite-bead behavior in CAD In this circuit, the overall effect due to parasitic coupling
software tools. To overcome this drawback, additional data in between layers is represented by the impedance
the form of input impedance measurements are typically
provided by the manufacturers. However, how these data were
 Fig. 1. Physical structure of SMT ferrite beads [1].

ZL
Fig. 3. Impedance-frequency characteristics of the ferrite bead
L( f ) R( f ) BLM18PG221SN1, as assigned by the manufacturer [2].
A B
where L0 is related to the bead geometrical layout, and
 ( f ),  ( f ) to the magnetic/dielectric properties of the
C
ZC ferrite the sheets are made of.
In typical ferrite materials used for SMT ferrite-bead
Fig. 2. Lumped-parameter circuit model of the SMT ferrite bead [1]. manufacture [4], the term  ( f ) prevails at low frequencies,
whereas at high frequencies  ( f ) is the larger contribution.
Z C   j /( C ) , where the inter-winding capacitance C takes
Accordingly, the component is expected to exhibit inductive
typical values on the order of C  0.4 ÷0.6 pF [1]. Conversely, behavior only in the low frequency range. Conversely, in the
the series impedance Z L is given by 100’s of MHz range, R ( f ) should be considered the
dominant mechanism for SMT ferrite bead impedance.
Z L  j Lˆ ( f )  R( f )  j L( f ). (1) This qualitative interpretation well agrees with the
frequency behavior of the impedance of typical SMT ferrite
It accounts for energy storage, L( f ) , and power losses, beads. As an example, Fig. 3 shows the impedance frequency
behavior of an BLM18PG221SN1 ferrite bead, measured by
R ( f ) , occurring within the ferrite sheets at high frequencies.
the component manufacturer and reported in [2]. The
Both R ( f ) and L( f ) are frequency-dependent parameters inductance of the bead is dominant only at lower frequencies
which can be interpreted by the light of the concept of (i.e., for frequencies below 25 MHz), whereas in the interval
complex permeability [9]. Indeed, while at low frequencies 25 MHz < f < 400 MHz, the resistive or dissipative term
the magnetic field and the auxiliary magnetic field are simply prevails over the reactive contribution. Finally, for frequencies
proportional to each other through some scalar permeability above 400 MHz, the impedance of the SMT ferrite component
(linear materials), at high frequencies these quantities will is mainly determined by parasitic capacitances.
react to each other with some lag time [10]. Therefore, if the However, in order for the model in Fig. 2 to accurately
permeability is expressed as the sum of a real and an predict the component impedance, knowledge of the
imaginary part as frequency behavior of R ( f ) and L( f ) is required. However,
in several application notes published by ferrite manufacturers,
ˆ ( f )   ( f )  j ( f ) , (2) ferrites are usually specified at 100 MHz, and values for the
R-L-C circuit elements in Fig. 2 are assigned at this specific
the complex inductance Lˆ ( f ) in (1) takes the form frequency. For the specific chip selected for this investigation
(BLM18PG221SN1 [1]), the following values were captured
(at 100 MHz) by using Agilent 4294A impedance analyzer:
Lˆ ( f )  L0  ( f )  j ( f ), (3) L  0.28 μH; R  220 ; C  0.5 pF. Resorting to these
values for circuit simulation leads to the impedance prediction
and yields the following expressions for R ( f ) and L( f ) : illustrated in Fig. 4, which reveals significant deviations with
respect to the impedance characteristics as illustrated in Fig. 3
L( f )   ( f ) L0 , (4.a) (note that the y-scale of the two figures is different, due to the
extremely large value reached by the predicted impedance at
the resonance frequency).
R( f )    ( f ) L0 , (4.b)
This confirms the inadequacy of a modeling strategy based
on the use of frequency-independent parameters, and
 3000
Z
R
2500 X

2000
Impedance [ ]

1500

1000

500
Fig. 5. Modified SMA receptacle for impedance characterization.
0 6 7 8 9
10 10 10 10
Frequency [Hz] ZF
Lc Lc
ZL
Fig. 4. Impedance of the ferrite bead BLM18PG221SN1 predicted by the
model in Fig. 2, by assuming the following frequency-independent values (at ZC
R( f ) Zm
100 MHz) for R, L, and C: L  0.28 μH ; R  220  ; C  0.5 pF. Cc
L( f ) C
strengthens the need for accurate knowledge of R ( f ) and
L( f ) in order to provide an insight for the accurate behavior SMT ferrite bead SMA receptacle

over the large frequency range. Fig. 6. Circuit interpretation of the impedance, Z m , measured at the port of
the SMA receptacle in Fig. 5.
III. MODEL EXTRACTION FROM MEASUREMENT DATA
In this section, the SMT ferrite bead (BLM18PG221SN1) is 1
Cc  . (6)
experimentally characterized via Vector Network Analyzer 2 f | Z inopen |
(VNA) measurements, with the objective of retrieving the
frequency-dependent behavior of R ( f ) and L( f ) from In a similar fashion, the inductance Lc is obtained from the
measured data. Therefore, the component was mounted on a input impedance measured with the receptacle loaded by a
“modified SMA receptacle”. A modification (i.e., stubs
short-circuit, Z inshort , as
elimination) was required to an SMA receptacle as illustrated
in Fig. 5 to eliminate, or reduce, the external parasitic and stub
effects. | Z inshort |
Lc  . (7)
Measurements were carried out via an Agilent (Hewlett 4 f
Packard) 8753E RF VNA in the frequency interval 1 MHz ÷ 2
GHz. The impedance Z m was obtained from the scattering The approximate expressions in (6), (7) are valid as long as
parameter S11 measured at the port of the modified SMA the operating frequency satisfies the inequality
receptacle as
1
f  . (8)
Z m  R0 (1  S11 ) /(1  S11 ) , (5) 2 Lc Cc

where R0  50  denotes the reference resistance used to Hence resulting, Cc  0.72 pF , Lc  0.97 nH .
define the scattering parameters. According to the above model of the SMA-receptacle, the
impedance of the SMT ferrite bead (left block in Fig. 6) is
A. De-embedding of connector-related effects
extracted from the impedance Z m as
In order to extract R ( f ) and L( f ) of the ferrite bead, the
spurious effects associated with the SMA receptacle must be
Z Lc ( 2 Z Cc  Z Lc )  Z m ( Z Cc  Z Lc )
de-embedded from the measured impedance Z m , at the SMA- ZF  , (9)
Z m  Z Lc  Z Cc
receptacle port of the VNA. The lumped-Tee circuit network
(right box in Fig. 6) is assumed for the SMA connector model
[11], and specific VNA measurements are used to determine where: Z Cc   j ( 2 f Cc ) 1 ; Z Lc  j 2 f Lc . Magnitude, real
the spurious circuit elements Lc , C c . and imaginary parts of the impedance Z F are plotted in Fig. 9
The capacitance C c can be extracted from the impedance as black lines.
open
Z in measured with the receptacle configured as open-ended, B. Evaluation of R( f ) and L( f )
i.e.,
Once the impedance of the SMT receptacle is determined, Z L
in (1) is readily evaluated by eliminating from Z F the effects
 250
|ZL|
Re(Z L)
200 Im(Z L)
Impedance Z L []

150

100

50
(a) (b)
0 6 7 8 9 Fig. 9. SPICE schematic of model no. 1 (a) and no. 2 (b) described in Sub-
10 10 10 10
Frequency, [Hz] sections IV.A and IV.B, respectively.
Fig. 7. Frequency behavior of impedance Z L extracted from measurement
1500
data as per (10): Magnitude (solid), real (dashed), and imaginary (dotted) part. LF @ 100 MHz
LF @ 1 MHz
2 1000

Input impedance [ ]
500
1.5
Inductance [ H]

0
1
-500 Abs
Real
0.5 Imag
-1000 6 7 8 9
10 10 10 10
Frequency, [Hz]
0 6 7 8 9
10 10 10 10 Fig. 10. SPICE model no. 1: Comparison between prediction and
Frequency, [Hz]
measurement (black curves) of impedance Z L . Red and green curves were
Fig. 8. Frequency behavior of inductance L( f ) . obtained for values of LF extracted from Fig. 8 at 1 MHz and 100 MHz,
respectively.
due to the inter-winding capacitance C  0.5 pF , resulting
Accordingly, impedance Z L results from the series
Z F ZC connection of (a) a constant-valued inductance, LF ; and (b) a
ZL  . (10)
ZC  Z F frequency-dependent resistance R( f ) . The SPICE schematic
is shown in Fig. 9(a). In the model, R ( f ) is implemented by
Magnitude (solid), real (dashed) and imaginary (dotted) a GFREQ module with trans-conductance g m ( f )  1 / ( Z L ) .
parts of impedance Z L are plotted versus frequency in Fig. 7.
Regarding LF , different values of this inductance are
Resistance R ( f ) directly coincides with the real part of Z L .
retrieved from Fig. 8, and used for SPICE simulations.
Inductance L( f ) is obtained as the imaginary part of Z L Magnitude (solid lines), real (dotted lines), and imaginary
divided by  . This leads to the frequency behavior illustrated (dashed lines) part of the impedance Z L predicted by the
in Fig. 8. The significant reduction of L( f ) at higher above model are compared against the measured data (black
frequencies is ascribed to magnetic dispersion phenomena. curves) in Fig. 10. In the first test-case (red curves), the value
LF  1.8 μH obtained from Fig. 8 at 1 MHz is adopted for
IV. SPICE IMPLEMENTATION simulation. The second test-case (green curves) makes use of
Model implementation in SPICE is achieved by resorting to the value LF  0.12 μH extracted from Fig. 8 at 100 MHz
ABM models [6, 7, 12]. Particularly, the frequency (note that the inductance measured at the same frequency, i.e.,
dependence of impedance Z L in (10) is included into the LF  0.28 μH , takes a quite larger value than LF ).
model by a GFREQ part, that is by means of a voltage- The comparison reveals non-negligible errors in the
controlled current-source with trans-conductance, gˆ m ( f ), reconstruction of the impedance Z L . In particular, while LF
assigned frequency-by-frequency by a look-up table [12]. Two yields a good prediction at extremely low frequencies, it
possible implementations of the ferrite-bead model will be introduces significantly high resonance peak approximately at
discussed in the following subsections. 200 MHz, with errors higher than 30 dB. Conversely, LF
A. SPICE model no. 1 assures better reconstruction of | Z L | in a limited interval
Since R ( f ) dominates at HF, in the first implementation (approximately between 30 MHz and 200 MHz). Finally, the
the frequency dependence of L( f ) is disregarded. value LF  0.28 μH , measured at 100 MHz, was used. Even if
omitted in Fig. 10, such a prediction is worse than those
 10

-10

-20

S21 [dB]
-30

-40

-50
Abs
-60 Real
Imag
-70 6 7 8 9
10 10 10 10
Fig. 11. Geometrical layout (upper view) of the PCB board used for Frequency, [Hz]
experimental characterization of the component in terms of IL.
Fig. 12. Magnitude (solid), real (dotted) and imaginary (dashed) parts of the
obtained by LF , since the resonance moves to lower scattering parameter S21 measured at the ports of the PCB board in Fig. 11.
Black curves represent measurement data, green lines are predictions obtained
frequencies (i.e., from  600 MHz to  400 MHz) and shows by the SPICE model in Fig. 13(b).
a sharper and larger peak ( 69 dBΩ ).
A. SPICE modeling of the test board
B. SPICE model no. 2 Accurate prediction of the above quantity ( S 21 ) requires
Previous results demonstrate the need for retaining the preliminary identification and modeling of the effects which
frequency-dependence of L( f ) in order to achieve accurate do not pertain to the SMT component. These include: a)
prediction in the wide-frequency range. This is accomplished propagation along the traces, b) mismatching and field
by the SPICE model illustrated in Fig. 9(b). In this model, the transitions at the interfaces with the SMA receptacles, c)
GFREQ module is used to include the frequency behavior of effects due to the SMA connectors.
the impedance Z L  R( f )  jL( f ) extracted from the In line with this need, an auxiliary PCB test-board was
realized and experimentally characterized. The board layout is
measured data [as per (10)]. Accordingly, its gain is assigned
exactly the same as in Fig. 11, with the exception of the SMT
as gˆ m ( f )  1 / Z L . This solution has the potential to allow for component that in this case is absent. Such a board was
accurate reconstruction of the input impedance Z F up to 2 characterized at the SMA ports via VNA measurements in the
GHz, even if a limited number of data-samples is used to range 1 MHz ÷ 2 GHz, and modeled in SPICE by the circuit
assign the trans-conductance gˆ m ( f ) to the GFREQ part. network in Fig. 13(a).
In this model, the inner PCB trace is modeled by a 30 mm
Namely, SPICE simulations (not reported here for brevity)
long transmission line (TL) with characteristic impedance
proved that resorting to a set of 15 logaritmically-spaced data-
points can suffice to this purpose. Z TL  128  and line velocity vTL  1.7  108 m/s , [13, Ch. 4,
Sec. 2].
V. MODEL VALIDATION Propagation effects along the SMA connectors as well as
For model validation, a set of measurements independent of field transitions at the interface between the connectors and
those used for model-extraction is required. To this end, the the inner trace are modeled by the cascade connection of
simple test board in Fig. 11 was realized and experimentally elemental L-C sections, whose values are fitted from the
characterized by VNA measurement. The ferrite chip is measurement data. Finally, HF losses are included into the
mounted on the upper side with its terminals connected to model by frequency-dependent resistors, r ( f )  k f (with
PCB-mounted SMA receptacles by a straight trace section.
The ground plane lies on the opposite side of the board. Trace parameters extracted from measurements), connected to the
width is 0.3 mm. Trace high above ground is 1.6 mm. Trace end-points of the inner TL and implemented in SPICE by two
thickness is 1.4 mils (i.e., 1 ounce copper). Relative GLAPLACE ABM parts [12].
WO
permittivity of the FR4 substrate is  r  4.7. The total S 21 predictions (red lines in Fig. 14) are computed as two
distance between the SMA receptacles is 30 mm. Accordingly, times the voltage level picked-up across the rightmost resistor
each trace section is approximately 14.2 mm long, since the R1  50  . The comparison versus measurement data (black
chip case (0603, [2]) has length 1.6 mm. lines) is shown in Fig. 14.
Measurements were carried out in the frequency range 1 MHz
÷ 2 GHz by connecting the VNA ports to the SMA receptacles. B. IL prediction
The IL coincides with the scattering parameter S 21 (i.e., the The overall SPICE model of the test setup used for S 21
transmission coefficient) measured between the two SMA prediction in the presence of the SMT ferrite-bead (Fig. 11) is
ports. Magnitude, real and imaginary parts of S 21 are plotted shown in Fig. 13(b). It is obtained combining the auxiliary
versus frequency in Fig. 12 as black lines. board model [in 13(a)] with the ferrite bead model [in Fig.
8(b)].
 (a)

(b)
Fig. 13. SPICE models of the PCB test-boards used for S21 measurement in the absence (a) and in the presence (b) of the SMT ferrite-bead.

10 samples suffices for proper representation of the trans-

0
conductance of the SPICE behavioral model (GFREQ ABM
-10
part), and for accurate prediction of the ferrite bead behavior.
The effectiveness of the proposed modeling approach has
-20
been validated up to 2 GHz by realization and modeling of ad
[dB]

-30
hoc PCB test-boards and by characterizing the SMT
SWO
21

-40
component in terms of IL. Although the analysis here
-50
Abs
presented focuses below 2 GHz since the specific chip
-60 Real selected for the investigation starts losing its effectiveness
Imag
-70 7 8 9 from 1 GHz, in principle the proposed modeling approach can
10 10 10
Frequency, [Hz] provide accurate prediction also for ferrite beads characterized
Fig. 14. Magnitude (solid), real (dotted) and imaginary (dashed) parts of the by performance region in wider frequency intervals.
scattering parameter S21 at the ports of the auxiliary PCB-board. Black curves
represent measurement data, red lines are predictions obtained by the SPICE REFERENCES
model in Fig. 13(a). [1] muRata, Ferrite Manual SMD/Block Type EMFIL EMI Suppression
Filters. [Online]. Available: www.murata.com.
Prediction data for S 21 (green lines) are compared versus [2] ——, EMIFIL (Inductor type) Chip Ferrite Bead- BLM18P Series (0603
measurement (black curves) in Fig. 12. The comparison shows Size), Data Sheet, 2009. [Online]. Availabe: www.murata.com.
[3] K. Naishadham, "Experimental equivalent-circuit modeling of SMD
maximum discrepancies on the order of 2 decibels, thus inductors for printed circuit applications," IEEE Trans. Electromagn.
providing satisfactory validation of the modeling approach Compat., vol. 43, no. 4, pp. 557-565, Nov. 2001.
proposed in Section IV. It is worth noting that, even if the [4] L. Hill, "Ferrite beads for noise mitigation for EMC applications,"
proposed model neglects possible interactions between the presented at the SE Michigan IEEE EMC Spring Fest, Canton, Michigan,
May 2009.
SMT component and the PCB board, satisfactory IL [5] J. Fan, S. Luan, and J. L. Drewniak, "Including SMT ferrite beads in DC
predictions are obtained in the frequency band of interest. power bus and high-speed I/O line modeling," in Proc. 2001 IEEE Int.
Symp. Electromagn. Compat., Montreal, Que., Canada, Aug. 13-17,
VI. CONCLUSION 2001, pp. 336-339.
[6] C. Basso, “SPICE analog behavioral modeling of variable passive”,
This study has examined fundamental problems Power Electronics Technology, Apr., 2005, p. 58.
encountered in modeling SMT ferrite-bead components, as [7] F. Grassi, F. Marliani, and S. A. Pignari, "SPICE modeling of BCI
one of the most widely used frequency-dependent components probes accounting for the frequency-dependent behavior of the ferrite
found in the realm of EMI suppression category. The core," in Proc. XIXth General Assembly of International Union of Radio
Science (U.R.S.I.), Chicago, IL, USA, Aug. 7-16, 2008, Paper E07.6.
modeling approach here proposed is based on preliminary S- [8] M. Damnjanovic, G. Stojanovic, V. Desnica, L. Zivanov, R.
parameters measurement, with the SMT component mounted Raghavendra, P. Bellew, N. Mcloughlin, "Analysis, design, and
on a modified SMA receptacle. It has been shown that: a) characterization of ferrite EMI suppressors," IEEE Trans. Magnetics,
standard procedures resorting to simple R-L-C circuit models vol. 42, no. 2, pp. 270-277, Feb. 2006.
[9] E. C. Snelling, Soft Ferrites: Properties and Applications, Iliffe Books
with frequency-independent parameters are inadequate to Ltd., London, 1969.
accurate prediction on the wide frequency range, b) accurate [10] M. Getzlaff, Fundamental of magnetism, Berlin: Springer-Verlag, 2008.
de-embedding of connector-related effects is a required step. [11] C. Rostamzadeh, S. Connor, B. Archambeault, "Numerical and
Then, simple use of analog behavioral modules [12] available experimental investigation of power supply noise decoupling strategies
on single-sided printed circuit boards," in Proc. 2007 IEEE Symp.
in low-cost (student) versions of SPICE is sufficient to Electromagn. Compat., Honolulu, HI, USA, Jul. 9-13, 2007.
represent and include into the model the frequency behavior [12] Cadence-Orcad PSPICE user guide.
of the ferrite bead. [13] C. R. Paul, Introduction to Electromagnetic Compatibility, Wiley
Particularly, this analysis has shown that if the frequency Interscience, Second Edition, 2006.
dependence of both the resistive and the inductive parts of the
input impedance are retained, a limited number of data-

View publication stats